Network Content Monitoring System Based On Local Information Semantic Bias Recognition Algorithm

The purpose of Net-Monitor System is to mainly monitor the illegal information in the internet. This paper designs and relizes an algorithm about semantic orientation recognization based on local information in layer-classifacation system. This algorithm adopts Hidden Markov Model (HMM) to semantic orientation recognization by local information of key words. We need determine proper local information of key words for improving the results of recognization. This paper presents a method based on a single sentence and adopts mutual information to compute total information that the sentence provides for key words. The results of experiments indicate that the window size is proper indeed and adequate information is gained. Considering the practical requirements, the system adopts the method that [-8, +9] to those texts which have not signed for sentences and a whole sentence for others.This paper presents the classification according to the context based on HMM. we choose several samples size, and train several suits of parameters by partial derivative reverse transmitting. After getting these parameters, we test them in abundant truly samples, then we can choose the properly sample size and it meets the need of our system. The results of experiments indicate that the HMM based on the single sentence for classification can avoid the problems that the vectors are too sparsity and the precision is not high etc. Afer testing this model by true samples,the system behaves high classification Precision and Recall.